[fitsbits] 64-bit integers

[Thomas McGlynn]

William Thompson wrote:

.. The obvious implementation would be to use BITPIX=64 in
> primary HDUs and IMAGE extensions, and TFORM='nK' in binary tables.  
> 
> 
In IDL the MWR/RDFITS pair added long integer support in early 2002 I think.
The Java FITS classes I wrote have always supported long integers.  How
well tested these are is another story though.

I recall running into one problem with long integers in IDL.  The FXPAR
routines that they use typically return numeric values as doubles, but this can
lose precision  for 64 bit integers.  There is also the question of if/how
one supports 64-bit unsigned integers.

In both cases these implementations support exactly the syntax
Bill suggested, but it may well be
that the single most important role of long integers is to have
longer values in the variable length array pointers.  Long values
would be particularly valuable for the offset value.  It can be
quite natural to have byte offsets in the heap that are larger than 2 GB.
Presumably we could use 64 bit integer pointers with
TFORM='Q' in binary tables but I have not implemented this.

	Regards,
	Tom McGlynn